WorkLink Create: Object Tracking
Please start by watching this video to have a full understanding of what object tracking is, how it works, what all settings mean and best practices to set up object tracking correctly:
See this article to see the easiest and fastest way to create Object trackers.
An explanation of what all settings do:
Inlier ratio tracking initiation
The inlier ratio tracking initiation setting sets the minimum inlier ratio required before tracking can initiate.
Setting this value too low, can result in false positive tracking. Setting it too high, might result in not being able to initiate the tracking at all. We recommend using a value between 0.4 and 0.7. Reaching inlier ratio values of 0.7 to 0.8 during the experience, are generally considered as a good result.
Inlier ratio tracking minimum
After tracking is initiated, the "inlier ratio tracking minimum" setting becomes relevant, it means that active tracking will stop if the inlier ratio drops below this value.
Laplace threshold
The Laplace threshold is an expression in millimeters. A line will be generated whenever the depth jump between surfaces from the camera's point of view, is larger than the defined value. A very high value will create a result that primarily shows the contours of your object, as the contours are basically an infinite depth jump.
Normal threshold
The normal threshold creates lines determined by changes in the surface direction. A high value will generate lines for hard edges, while a lower value will also generate lines for softer edges. Generally we do not recommend setting this value below 0.1.
Key frame distance
The minimum distance between keyframes in mm. The line-model will only be generated for certain keyframes. Therefore a higher value improves the performance at the cost of a lower precision. Lower values cost more performance but increase the precision.
Line gradient threshold
Threshold to determine what is considered a contrast line in the video feed. Low values will consider low contrast areas as a candidate, which may result in false positives. High values will only consider very high contrast areas as a candidate, which may result in a lower or too low inlier ratio, so the tracking could fail to initiate. We recommend leaving this setting to the default.
Line search length during tracking
This setting determines the orthogonal length around the line model, in which contrast will be searched in the video feed, after tracking initiated.
Line search length before tracking
This setting determines the orthogonal length around the line model, in which contrast will be searched in the video feed, before tracking initiates.
Active Tracking Visibility Threshold
Sets the minimum percentage of a 3D model that must be visible to the device’s camera for active tracking to override SLAM-based positioning.
The reason to use this feature can be for cases when getting up close to the model, resulting in a low amount of lines to be visible to the camera, resulting in possible false positives, which could make the 3D model to track unstable. Reverting to SLAM-based positioning in such cases can result in better stability.
The default setting is 60%, but you can experiment by setting this value lower, eg. 10% for line models that have sufficient unique features even while being zoomed in.
Fig. 1: The 3D object is tracking and completely in view. In debug mode, you can observe the "Visible Percent" is currently 91.6%
Fig. 2: The 3D object is only 40.7% in view after tracking started. As the author set the "Active Tracking Visibility Threshold" to be higher than 60% in this case, this value displays yellow, to indicate the user that tracking switched from active tracking to SLAM-based positioning.
When testing these settings, we recommend using debug mode in the app, to validate if the settings you use are the best for your use case.
📘 Mastering Object Tracking: A Guide to Precision Settings
Object Tracking works by comparing a 3D Line Model (the digital skeleton of your object) against high-contrast edges in the real-world video feed. To get a "lock," the software looks for a specific percentage of matching lines, known as the Inlier Ratio.
🔍 How the Tech Sees Your World
Search: The app looks within a defined area (Search Length) for contrast.
Verify: It checks if the contrast lines match the shape and surface direction (Normal/Laplace) of your 3D model.
Lock: If enough lines match (Inlier Ratio Initiation), the 3D content snaps into place.
Maintain: As you move, it uses SLAM (Spatial Mapping) to keep the object stable even if you look away briefly.
🛠️ Troubleshooting & Optimization Table
Use this table to diagnose tracking issues seen during testing in the WorkLink App Debug Mode.
| If you see... | The Cause is likely... | Adjustment Strategy |
|---|---|---|
| Too many red lines on a shiny/metallic object. | Reflections are creating "fake" edges that don't exist on the 3D model. | Increase Laplace Threshold. This forces the app to only look for deep physical edges, ignoring surface glares. |
| No lines at all on a complex or curved part. | The "edge" is too soft for the current sensitivity. | Decrease Normal Threshold. This makes the app more sensitive to subtle changes in surface direction. |
| Tracker "snaps" to the wrong object or background clutter. | The initiation settings are too "desperate" to find a match. | Increase Inlier Ratio Initiation and Decrease Line Search Length. This tightens the "search grid." |
| Tracking "pops" or jumps when moving the camera. | The system is searching too far from the last known position. | Decrease Line Search Length during tracking. This anchors the model more strictly to the visual edges. |
| Tracking stops when you get close to the object. | Not enough of the model is visible to meet the "Confidence" check. | Lower the Active Tracking Visibility Threshold. This allows the app to rely on SLAM when you are zoomed in. |
| Heavy jitter on large objects (over 2m). | Small camera tremors are fighting against a massive coordinate system. | Increase Keyframe Distance. This reduces the frequency of "re-calculating" the model position, smoothing the experience. |
💡 Advanced Authoring Tips
1. The "Perfect" 3D Candidate
Unique Geometry: Avoid perfectly symmetrical objects (like a plain pipe or a sphere). The tracker needs a "unique hook" (a notch, a handle, a specific bolt pattern) to know which way is forward.
High Contrast: Tracking a dark gray machine in a dark gray factory is difficult. If possible, choose tracking targets that have a different color/value than the floor or walls.
2. Environmental Control
Avoid "Flat" Lighting: Ironically, very bright, overhead fluorescent lights can "wash out" edges. Soft, directional light that creates clear shadows on the object’s edges is ideal.
Manage Specular Highlights: On shiny objects (Chrome, Polished Steel), try to stand so the main light source isn't bouncing directly into the lens. These "hot spots" look like edges to the software and will turn your debug lines red.
3. Strategic "Active Tracking Visibility"
If you are tracking a large aircraft engine but the user needs to inspect a small fuel valve:
Set the Visibility Threshold low (e.g., 10-20%).
This tells the app: "Once you've found the whole engine, keep it locked even if the user zooms in so close they can only see the valve."